Research On Citation Recommendation Method Of Scientific And Technological Literature Based On Deep Learning

Science and technology literature resources are the main manifestation of the results of scientific and technological innovation activities,and an important carrier for the dis-semination of scientific and technological knowledge.They are also the basic source and important support for further enhancing the ability of scientific and technological innova-tion.They have become one of the country's precious strategic resources.In the process of scientific and technological innovation,it is particularly important for the innovation subject to understand the development history and trends of the disci-pline,and effective academic communication with peers,and consulting related scientific and technological literature in the field is the best way to achieve this process.However,whether the scientific and technological literature obtained is comprehensive and whether the content is relevant will directly affect the effectiveness of innovation.Therefore,it is of practical significance and great demand to support the research of scientific and tech-nological literature acquisition methods by innovative entities.The existing scientific and technological literature acquisition method is mainly the"keyword" retrieval method,but this method is often affected by human factors and re-trieval tools,and it is often difficult to meet the requirements of document coverage and accuracy.Because of the huge number of scientific and technological documents,the va-riety of them,scattered and isolated,dynamic heterogeneous,diverse and complex,and have extremely strong professional,academic and unstructured characteristics.Existing methods are difficult to guarantee the accuracy of the"keyword" setting due to their de-pendence on human factors,which will inevitably affect the efficiency of search and the comprehensiveness and accuracy of the results.In addition,due to the lack of semantic recognition and matching reasoning capabilities of retrieval tools based on the "keyword"method,they cannot support the semantic understanding and recognition and matching of document content,which will also affect the comprehensiveness and accuracy of docu-ment retrieval.In order to solve the problem that the traditional "active search" method is overly dependent on human factors and the search tools lack matching reasoning ability,the text reasoning matching and citation recommendation methods that support the acqui-sition of scientific literature have become an urgent need and a hot research direction.To this end,this article uses the national key research and development program topics" Support for the development and application of an open and ecological enterprise-level cloud ERP platform"(topic number:2019YFB1704104)and "distributed resource giant system and resource synergy theory"(topic number:2017YFB1400301)For the re-search background,focus on the subject of data-driven intelligent service technology to enhance the goal of integrating science and technology resources into the industry,ex-plore the sharing and service mode of science and technology resources,and the Wanfang Science and Technology Service Platform,Ningbo Science and Technology Information Research Institute and Dongfang Lingdun as required by the task.The document resources in the patent service platform are data support,and are oriented to document acquisition applications that support scientific and technological innovation.In response to the exces-sive dependence on human factors in the traditional "active search" retrieval method,and the lack of matching reasoning capabilities in retrieval tools,the development of support for the acquisition of scientific and technological documents Research on text reasoning matching and citation recommendation methods.The specific content is as follows:(1)Analyze the needs of scientific and technological text-oriented matching and cita-tion recommendation,analyze the characteristics of scientific and technological literature text resources,as well as the current situation and problems of citation recommendation,and form an overall plan for scientific and technological literature citation recommenda-tion based on deep learning.The learned text reasoning matching scheme and the two-stage citation recommendation scheme based on text content are composed of two parts,which respectively solve the semantic understanding,content completeness and accuracy problems of the existing scientific and technological literature acquisition methods in text reasoning and matching.(2)According to the features of high-noise,large-scale professional vocabulary,and unstructured texts of scientific and technological literature,preprocessing such as denois-ing,word segmentation,and removal of stop words is carried out on scientific and tech-nological texts.In order to better represent the semantic features of scientific and techno-logical texts,word2vec technology is used to train distributed word vector expressions of texts,providing high-quality semantic features and digital text input forms for subsequent text inference matching and citation recommendation.(3)Aiming at the problems of low accuracy of existing scientific and technological text inference matching methods,the need to manually extract features,and the difficulty in understanding text semantics with existing statistical methods,a text inference match-ing method based on LSTM and CNN is proposed.This method first uses Bi-LSTM to represent the text to be matched by semantic vectorization,then uses the Attention mech-anism to interactively encode the text,and finally uses the multi-layer CNN network to extract the interactive information of the text to obtain the text Final match information.Through experiments,it is verified that the method in this paper has better results.(4)The current citation recommendation method is based on the source data and ci-tation relationship network,which causes the lack of semantic information of the text con-tent.A two-stage citation recommendation method based on text content is proposed.The first stage uses the vector space similarity of the text to generate a related citation recom-mendation set,and the second stage uses a text inference matching method to understand the language of the candidate set to obtain more accurate relevance Sorted list of degrees.The experimental results show that the method in this paper has a better recommendation effect.(5)In order to verify the applicability of the method in this paper,part of the data on the science and technology resource service platform is selected for actual verification.After comparing with the existing methods,the method proposed in this paper has better results and verifies the feasibility of the method in this paper.